File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Conference Paper: Online Knowledge Distillation via Collaborative Learning

TitleOnline Knowledge Distillation via Collaborative Learning
Authors
KeywordsKnowledge engineering
Collaborative work
Perturbation methods
Learning (artificial intelligence)
Neural networks
Issue Date2020
PublisherIEEE Computer Society. The Journal's web site is located at http://ieeexplore.ieee.org/xpl/conhome.jsp?punumber=1000147
Citation
Proceedings of IEEE/CVF International Conference on Computer Vision and Pattern Recognition (CVPR 2020), Seattle, USA, 13-19 June 2020, p. 11017-11026 How to Cite?
AbstractThis work presents an efficient yet effective online Knowledge Distillation method via Collaborative Learning, termed KDCL, which is able to consistently improve the generalization ability of deep neural networks (DNNs) that have different learning capacities. Unlike existing twostage knowledge distillation approaches that pre-train a DNN with large capacity as the “teacher” and then transfer the teacher’s knowledge to another “student” DNN unidirectionally (i.e. one-way), KDCL treats all DNNs as “students” and collaboratively trains them in a single stage (knowledge is transferred among arbitrary students during collaborative training), enabling parallel computing, fast computations, and appealing generalization ability. Specifically, we carefully design multiple methods to generate soft target as supervisions by effectively ensembling predictions of students and distorting the input images. Extensive experiments show that KDCL consistently improves all the “students” on different datasets, including CIFAR100 and ImageNet. For example, when trained together by using KDCL, ResNet-50 and MobileNetV2 achieve 78.2% and 74.0% top-1 accuracy on ImageNet, outperforming the original results by 1.4% and 2.0% respectively. We also verify that models pre-trained with KDCL transfer well to object detection and semantic segmentation on MS COCO dataset. For instance, the FPN detector is improved by 0.9% mAP.
DescriptionSession: Oral 3.2A — Recognition (Detection, Categorization) (2) - Poster no. 5; Paper ID 6687
CVPR 2020 held virtually due to COVID-19
Persistent Identifierhttp://hdl.handle.net/10722/284162
ISSN
2023 SCImago Journal Rankings: 10.331

 

DC FieldValueLanguage
dc.contributor.authorGuo, Q-
dc.contributor.authorWang, X-
dc.contributor.authorWu, Y-
dc.contributor.authorYu, Z-
dc.contributor.authorLiang, D-
dc.contributor.authorHu, X-
dc.contributor.authorLuo, P-
dc.date.accessioned2020-07-20T05:56:34Z-
dc.date.available2020-07-20T05:56:34Z-
dc.date.issued2020-
dc.identifier.citationProceedings of IEEE/CVF International Conference on Computer Vision and Pattern Recognition (CVPR 2020), Seattle, USA, 13-19 June 2020, p. 11017-11026-
dc.identifier.issn1063-6919-
dc.identifier.urihttp://hdl.handle.net/10722/284162-
dc.descriptionSession: Oral 3.2A — Recognition (Detection, Categorization) (2) - Poster no. 5; Paper ID 6687-
dc.descriptionCVPR 2020 held virtually due to COVID-19-
dc.description.abstractThis work presents an efficient yet effective online Knowledge Distillation method via Collaborative Learning, termed KDCL, which is able to consistently improve the generalization ability of deep neural networks (DNNs) that have different learning capacities. Unlike existing twostage knowledge distillation approaches that pre-train a DNN with large capacity as the “teacher” and then transfer the teacher’s knowledge to another “student” DNN unidirectionally (i.e. one-way), KDCL treats all DNNs as “students” and collaboratively trains them in a single stage (knowledge is transferred among arbitrary students during collaborative training), enabling parallel computing, fast computations, and appealing generalization ability. Specifically, we carefully design multiple methods to generate soft target as supervisions by effectively ensembling predictions of students and distorting the input images. Extensive experiments show that KDCL consistently improves all the “students” on different datasets, including CIFAR100 and ImageNet. For example, when trained together by using KDCL, ResNet-50 and MobileNetV2 achieve 78.2% and 74.0% top-1 accuracy on ImageNet, outperforming the original results by 1.4% and 2.0% respectively. We also verify that models pre-trained with KDCL transfer well to object detection and semantic segmentation on MS COCO dataset. For instance, the FPN detector is improved by 0.9% mAP.-
dc.languageeng-
dc.publisherIEEE Computer Society. The Journal's web site is located at http://ieeexplore.ieee.org/xpl/conhome.jsp?punumber=1000147-
dc.relation.ispartofIEEE Conference on Computer Vision and Pattern Recognition. Proceedings-
dc.rightsIEEE Conference on Computer Vision and Pattern Recognition. Proceedings. Copyright © IEEE Computer Society.-
dc.rights©2020 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.-
dc.subjectKnowledge engineering-
dc.subjectCollaborative work-
dc.subjectPerturbation methods-
dc.subjectLearning (artificial intelligence)-
dc.subjectNeural networks-
dc.titleOnline Knowledge Distillation via Collaborative Learning-
dc.typeConference_Paper-
dc.identifier.emailLuo, P: pluo@hku.hk-
dc.identifier.authorityLuo, P=rp02575-
dc.description.naturelink_to_subscribed_fulltext-
dc.identifier.doi10.1109/CVPR42600.2020.01103-
dc.identifier.scopuseid_2-s2.0-85094605661-
dc.identifier.hkuros311022-
dc.identifier.spage11017-
dc.identifier.epage11026-
dc.publisher.placeUnited States-
dc.identifier.issnl1063-6919-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats